You have twenty years of clinical experience and hundreds of cases treated, but when a patient asks ChatGPT who to contact for your kind of specialty, the AI cites anonymous blogs written by nobody. In fields like medicine, AI looks for precise signals of authority and regulatory compliance — it doesn't find them on you, and it skips right past. Every patient who relies on that answer ends up somewhere else. Making those signals readable is simpler than it seems.
You run a specialist medical practice that invests in quality content, has an up-to-date website, three doctors with scientific publications and twenty years of case experience. Then someone opens ChatGPT and asks “best orthopedic specialist in Ancona” or “experienced cardiologist in Le Marche”: the AI answers by citing anonymous generalist blogs, online booking portals, sites stuffed with SEO content written by copywriters who know as much about medicine as you do about cardiology.
Your brand doesn’t show up. Ever.
It’s not a bug, it’s not bad faith, it’s not that your site is slow. It’s that in YMYL sectors — Your Money Your Life, meaning healthcare, legal, finance — AI models apply a stricter entity verification filter than in other sectors. And if your credentials aren’t structured in a way the machine reads as credibility signals, you get excluded. Not rejected: excluded upstream, for safety.
Let me explain how the filter works, what to publish so you pass it, and let me tell you the real case of an orthopedic practice in Le Marche that in four months went from zero citations to stable visibility in AI answers.
What the machine sees when it reads your practice’s page
In the world of research on credibility assessment with language models, Ivan Srba and his group published a survey describing how AI systems evaluate the reliability of a source before citing it. The process has two phases.
“First, granular credibility signals (also referred to as credibility indicators) are detected.”
Translated: the system first scans the page looking for granular credibility signals — specific, precise, verifiable elements. It doesn’t judge the tone, it doesn’t judge the length of the text: it looks for precise markers like a proper name, a professional registry number, an academic qualification, an institutional affiliation.
The operational consequence for you is clear. If your “About us” page says “Dr. Rossi, orthopedist with twenty years of experience”, the machine reads a generic string. If instead it says “Dr. Mario Rossi, registered with the Medical Board of Ancona no. 12345, specialization in Orthopedics and Traumatology at the Marche Polytechnic University (2003), member of SIOT”, the machine reads five separate, verifiable signals. In the first case you get discarded for lack of evidence; in the second you pass the first filter.
Why YMYL sectors have a stricter filter
From the same survey comes the reason why healthcare, legal and finance are treated with more rigor than, for example, tourism or interior design.
“False content that is perceived as highly credible by its audience can lead to significantly greater harm than false content that is clearly perceived as lacking credibility.”
In plain terms: false content that looks credible does far more damage than false content you immediately recognize as garbage. This is the principle that pushes ChatGPT, Claude, Gemini and Perplexity to raise the bar in three specific areas: if they suggest the wrong cardiologist, an incompetent lawyer or a phantom financial advisor, the harm to the user is direct and measurable. So better to cite no one than to cite badly.
Translated into practice for your practice: in your sector the bar is high by design. It’s not a prejudice against small businesses, it’s a safety mechanism. And you clear it in one way only: by giving the machine enough verifiable signals to make the recommendation worth the risk.
If you want to understand more deeply how AI engines build the reputation of an author or an organization, I covered the issue in the articles on E-E-A-T for AI and author entity recognition: they are the context everything you’re reading here rests on.
Professional registry number missing or hidden in the footer.
Here’s a test you can run in ten minutes
Before spending money on consultants, run three checks yourself.
First: open Google’s Rich Results Test. Paste the URL of the practice’s “About us” page and look in the results for the entities “Person”, “Physician”, “MedicalBusiness” or “Organization”. If you find nothing, the site isn’t telling the machine that there are structured professionals with credentials. Binary threshold: either the schema is there, or it isn’t.
Second: open displaCy ENT and paste the text of your bio page. Look at how many PERSON, ORG and GPE (places) entities are recognized automatically. If on the biography of a doctor with twenty years of career the system extracts three blurry entities, your page is opaque to machines.
Third: open Wikidata and search for the name of your practice and of the individual professionals. If no records exist, there’s no anchor. AI models use Wikidata as a reference graph to disambiguate entities with the same name.
With these three checks you’ll understand within half an hour whether your practice is invisible for structural reasons or for content reasons. It’s an entry-level audit, to be clear: real analysis requires professional tools and continuous editorial work. But it gives you an honest first snapshot.
Make sure every professional in the practice has a dedicated page with: full name, registry number, registering board, university of graduation, specializations with year, scientific societies they belong to.
The case of the orthopedic practice in Le Marche
Four specialists, based in Ancona, a solid body of cases, a well-kept WordPress site, a blog updated twice a month with technical articles written by the doctors themselves. When we started, ChatGPT and Perplexity cited the practice zero times for the query “orthopedist Ancona” and variants. They cited two booking portals, a medical-information blog based in Rome and a non-specialist private clinic.
The intervention was surgical, nothing magical.
First: adding `Physician` schema for each of the four specialists, with `identifier` fields (registration with the Ancona Board), `alumniOf` (university of graduation and specialization), `memberOf` (scientific societies: SIOT, SIAGASCOT), `knowsAbout` (specific clinical subspecialties), `medicalSpecialty`. `MedicalBusiness` schema for the practice with structured address, VAT number, hours, accreditations.
Second: rewriting the biographies. Out with phrasing like “extensive experience in knee surgery”: in with numbers, years, documented procedures, publications with DOI, conferences with links.
Third: creating Wikidata records for the two specialists with the most publications indexed on PubMed, anchoring the person entity to the practice entity and to the university/scientific-society entities.
Result after four months, measured on a sample of eighty real queries in the sector (produced by varying specialization, area, symptom): citation of the practice in nineteen cases on Perplexity, eleven on ChatGPT, eight on Gemini. From zero to stable presence in about a quarter of the relevant queries. An indicative test, not a huge sample, but a clear and replicable pattern.
The mistakes I see most often
When I analyze medical, legal or financial practices that don’t show up in AI answers, four patterns repeat.
Self-referential biographies with no verifiable data. “Dr. X is a reference point in the field” is not a signal, it’s an opinion. The machine doesn’t process it.
Professional registry number missing or hidden in the footer. It’s the strongest signal of professional credential in your sector, and you often treat it as a bureaucratic detail instead of as the most important verification marker.
Generic “LocalBusiness” schema markup instead of “Physician”, “Attorney”, “FinancialService”. The specific type carries a set of dedicated fields that the generic one doesn’t have.
Zero distinction between the individual professionals and the practice as an entity. A practice with four specialists must have five entities: the practice plus the four doctors, linked to each other with explicit relationships. A single information blob “we are a practice of four specialists” is weaker.
The other side: why few really pass the filter
Let me close with another passage from Ivan Srba’s survey, which explains why the majority of Italian YMYL sites stay out.
“Notably, there is a scarcity of approaches that detect and aggregate multiple credibility signals simultaneously.”
Translated: at the moment few systems manage to detect and aggregate multiple credibility signals simultaneously. The practical consequence is that when AI models aggregate, they prefer few sources that accumulate many clear signals, rather than many sources that each accumulate few. So you don’t need to have everything perfect: you need to have enough, and to have them structured so they’re readable in the same pass.
The orthopedic practice in Le Marche I told you about had no citations in national outlets, no TED profiles, no published books. It had registry registrations, academic qualifications, scientific societies and about ten peer-reviewed publications. Well structured, they were enough.
What you can concretely do this week
- Make sure every professional in the practice has a dedicated page with: full name, registry number, registering board, university of graduation, specializations with year, scientific societies they belong to.
- Add sector-specific schema markup (Physician, Attorney, FinancialService) for each professional and MedicalBusiness/LegalService for the practice. Test it with the Rich Results Test.
- For the two/three professionals with the most publications or recognitions, create a Wikidata record and link it to the practice entity.
- Compare your “About us” page with that of the three or four competing practices the AI cites in your sector and your area: look at what they have that you’re missing.
- Don’t promise results in two weeks. Recognition by AI models happens over reindexing and graph-update cycles: it generally takes three to six months for entities to consolidate.
In YMYL sectors the credibility filter is not a penalty, it’s an entry threshold. Clearing it doesn’t guarantee you top positions in AI answers, but it’s the condition without which content work has no effect. It’s the ground everything else I tell you about in this series then rests on.
If this principle was useful to you, the natural next steps are the articles on periodic entity audit, local entity in your city’s knowledge graph and entity-to-entity relationship mapping: together they cover the ongoing work that keeps your practice visible in AI answers month after month, not just at the first go-live.
